code-explorer

name	code-explorer
description	Query and analyze codebases using the Code Explorer knowledge graph. Use for finding symbols, analyzing impact, understanding architecture, exploring code relationships, generating documentation, and asking questions about code.
argument-hint	[command] [arguments]
allowed-tools	Bash(code-explorer *), Bash(*/code-explorer *), Bash(cargo run -p code-explorer-cli -- *), Read, Grep, Glob

Code Explorer — Knowledge Graph Code Intelligence

You have access to Code Explorer, a code-intelligence tool that builds a knowledge graph from source code. Use it to answer questions about codebases, find symbols, trace dependencies, and analyze impact — using a fraction of the context that reading files would cost.

Binary location

The CLI is code-explorer. After cargo build --release the binary is at target/release/code-explorer; put it on your PATH, or call it by full path from other projects. Inside the Code Explorer repo, cargo run -p code-explorer-cli -- also works.

See the value (quick)

code-explorer demo [path]            # Measure the context an LLM agent saves vs reading files

Available commands

1. Analyze a codebase (must run first)

code-explorer analyze [path]              # Index a repository
code-explorer analyze [path] --force      # Force re-index
code-explorer analyze [path] --incremental # Only re-parse changed files
code-explorer analyze [path] --embeddings  # Generate ONNX semantic embeddings (feature gated)
code-explorer analyze [path] --skip-git    # Skip git history phases (required for non-git folders)
code-explorer status                       # Check if index exists (+ last index duration / phase timings)

2. Search the knowledge graph

code-explorer query "authentication middleware"    # Natural language search
code-explorer query "user service" --limit 5       # Limit results

3. Symbol context (360-degree view)

code-explorer context UserService          # Callers, callees, imports, exports, hierarchy
code-explorer context handleRequest --repo my-project

4. Impact analysis (blast radius)

code-explorer impact handleRequest --direction both       # Upstream + downstream
code-explorer impact UserService --direction upstream      # Who calls this?
code-explorer impact UserService --direction downstream    # What does this call?

5. Ask questions (LLM-powered, requires ~/.codeexplorer/chat-config.json)

code-explorer ask "how does the bareme calculation work?" --path /path/to/project
code-explorer ask "which controllers call the external API?"

6. Code health report

code-explorer report --path /path/to/project        # Text report with grade A-E
code-explorer report --path /path/to/project --json # JSON output

7. Git analytics (requires the target to be a git repo)

code-explorer hotspots --path [path]       # Most changed files (last 90 days)
code-explorer coupling --path [path]       # Files that change together
code-explorer ownership --path [path]      # Code ownership by author

8. Tracing coverage & dead code

code-explorer coverage                             # Global tracing + dead code stats
code-explorer coverage UserService                 # Single class coverage
code-explorer coverage --json                      # JSON output
code-explorer coverage UserService --trace         # Flow trace mode

9. Raw Cypher queries

code-explorer cypher "MATCH (n:Function) RETURN n.name LIMIT 10"
code-explorer cypher "MATCH (n:Controller)-[:DEFINES]->(a:ControllerAction) RETURN n.name, a.name"
code-explorer cypher "MATCH (n:Method) WHERE n.name STARTS WITH 'Get' RETURN DISTINCT n.name"
code-explorer cypher "MATCH (n:Function) WHERE n.name CONTAINS 'auth' OR n.name CONTAINS 'login' RETURN n"
code-explorer cypher "MATCH (n:Method) WHERE NOT n.filePath ENDS WITH '.test.cs' RETURN n.name"

Supported Cypher operators:

• WHERE: =, <>, !=, CONTAINS, STARTS WITH, ENDS WITH
• Logic: AND, OR, NOT (precedence: NOT > AND > OR)
• RETURN: DISTINCT, count()
• Clauses: ORDER BY [ASC|DESC], LIMIT

10. Interactive shell

code-explorer shell                           # REPL with tab completion
# Inside the shell: query, context, impact, cypher, hotspots, stats, help

11. MCP server (for AI agents)

code-explorer mcp                                  # Start MCP server (stdio, JSON-RPC 2.0)
# 30 tools, including: list_repos, query, context, impact, detect_changes, rename,
#   cypher, search_code, read_file, find_cycles, find_similar_code, hotspots, coupling,
#   ownership, coverage, diagram, report, analyze_execution_trace, get_complexity,
#   list_todos, list_endpoints, list_db_tables, list_env_vars, get_endpoint_handler,
#   get_insights, save_memory
code-explorer mcp-install                          # Auto-configure the MCP server for Claude Code

12. HTTP server (REST API)

code-explorer serve --port 3000                    # JSON-RPC /mcp + REST + SSE
code-explorer serve --port 3000 --host 0.0.0.0     # Expose on all interfaces

13. Validate LLM config

code-explorer config test                          # Check API key + test connection

14. List indexed repos

code-explorer list                                 # Show all indexed repositories

15. Generate documentation (all formats)

code-explorer generate context   --path [path]    # AGENTS.md at repo root
code-explorer generate wiki      --path [path]    # /wiki/*.md (one per module)
code-explorer generate docs      --path [path]    # .codeexplorer/docs/*.md
code-explorer generate docx      --path [path]    # documentation.docx (Word, full doc with TOC)
code-explorer generate html      --path [path]    # DeepWiki-style HTML site
code-explorer generate obsidian  --path [path]    # Obsidian vault export
code-explorer generate all       --path [path]    # All formats above in one run

Flags on every generate subcommand: --output-dir <dir>, --enrich (LLM, needs chat-config.json), --enrich-profile <fast|quality|strict>, --enrich-lang <auto|fr|en>, --enrich-citations.

16. Other commands

code-explorer trace-files ClassName            # All source files involved in a feature
code-explorer diagram ClassName --type flowchart|sequence|class
code-explorer rag-import <docs-folder> --path [path]   # Import .md/.docx specs, link to code
code-explorer watch [path]                     # Re-index on file changes (debounced)
code-explorer dashboard [path]                 # Interactive terminal UI over the graph
code-explorer clean [--force|--all]            # Delete an index
code-explorer setup                            # Configure editor MCP integration

How to use this skill

When the user asks about code structure, architecture, dependencies, or impact:

1. Check if the repo is indexed: run code-explorer status in the relevant directory.
2. If not indexed: run code-explorer analyze [path] first (add --skip-git if the target isn't a git repo).
3. Choose the right command based on the question:
   • "Where is X defined?" → code-explorer query "X" or code-explorer context X
   • "What calls X?" → code-explorer impact X --direction upstream
   • "What does X depend on?" → code-explorer impact X --direction downstream
   • "How healthy is the code?" → code-explorer report
   • "Which files change most?" → code-explorer hotspots
   • "Is this code used?" → code-explorer coverage or code-explorer coverage ClassName
   • "Explain how X works" → code-explorer ask "how does X work?"
   • "Show me the architecture" → code-explorer generate html then read the output
   • "Which files are involved in X?" → code-explorer trace-files X
   • "Generate a diagram of X" → code-explorer diagram X --type flowchart
4. Parse the output and present it clearly with file paths and relationships.
5. Combine multiple commands for complex questions.

Graph node types

The knowledge graph contains 50+ node types:

• Code: Function, Method, Class, Interface, Struct, Enum, Module, Namespace
• ASP.NET: Controller, ControllerAction, Service, Repository, ExternalService
• UI: UiComponent (Telerik/Kendo grids), ScriptFile, AjaxCall
• Data: Entity (EF6), Association, NavigationProperty
• Infrastructure: File, Directory, Import, Export
• RAG: Document, DocChunk (from rag-import)

Relationship types

• Calls, CallsAction, CallsService — invocation chains
• Imports, Exports — module dependencies
• Inherits, Implements — type hierarchy
• HasMethod, HasProperty — class membership
• BelongsTo, Mentions — RAG doc-to-code anchors

Tips

• The .codeexplorer/ directory in the repo root holds the serialized graph (graph.bin), meta.json, and metrics.json (index timing/throughput).
• Supports 14 languages: JS, TS, Python, Java, C, C++, C#, Go, Rust, Ruby, PHP, Kotlin, Swift, Razor.
• Skips obj/, bin/, node_modules/, packages/ during analysis.
• Use --skip-git for non-git folders; hotspots/coupling/ownership return empty there (expected).
• All --json flags output machine-readable JSON.